Many-to-many image registration for tiled multitemporal remote sensing image data

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Tiling is a common process for remote sensing image data. Tiling the remote sensing image data is beneficial, especially for streamlining many further processes. However, shifting may occur during the tiling process. Although the shift occurs only in subpixels, it is visually noticeable and might cause analysis errors in multitemporal data. In general, image registration is carried out between the data to be registered and the data that is used as a reference (one-to-one or one-to-many image registration). This approach is possible when there is definitive data that can be used as a reference. However, there is no definitive data that can be used as a reference when tiled data shifts, since all of the data are affected by the tiling process. In this study, many-to-many image registration is proposed. Multitemporal data are used to calculate the average of the pixel distances from paired tiles. The calculation result becomes the reference value for shifting the tiles so that the tiles are aligned. Landsat-8 scenes that were acquired in 2019 are tiled. Then the pixel distance between tiles in the same location is calculated. The shifts are calculated using the average of the pixel distances. Synthetic data are used to perform the initial assessment, where some tiles were shifted by some subpixels values and then shifted back using the proposed method. The same assessment method is then carried out for all the tiles generated from the Landsat-8 scenes. The results show the average effectivity of synthetics data is higher than 76%, while on the overall data test, the effectivity is higher than 51%.

This publication has 18 references indexed in Scilit: